5 min read·Updated 2026-04-10

Auto Mode

Goal-driven autonomous workflows with cron scheduling, a task state machine, and a structured diary for inspection.

Mental model

Auto Mode lets the agent pursue an objective across multiple triggers without a human in the loop each turn. The operator defines a goal, the agent breaks it into tasks, the scheduler triggers execution on a cron expression, and the diary records what happened for inspection.

The reader should picture Auto Mode as a cron-driven loop over a persistent task list: on each tick, the agent picks up the next pending task, executes it in a normal tool loop, updates state, and writes to the diary.

Core concepts

Goal

The high-level objective with its own status. Owns the task list and the diary. Persisted in workspace/auto-mode/.

Container

Task

A concrete step toward the goal. Has its own state machine (see below) and carries instruction text for the agent.

Unit of work

Diary

Structured log of what happened in each run: THOUGHT, PROGRESS, OBSERVATION, DECISION, ERROR entries.

Audit

Schedule

A cron expression attached to a goal or task. Triggers execution on a schedule.

Trigger

Task state machine

A task moves through five states. Transitions are explicit — either the agent moves a task via the goal_management tool, or the scheduler triggers a run that picks up the next pending task.

PENDING → IN_PROGRESS — a run picked up the task. The agent logs a THOUGHT entry and begins executing.
IN_PROGRESS → COMPLETED — the task finished successfully. The agent logs a PROGRESS entry and moves to the next pending task.
IN_PROGRESS → FAILED — the task hit an error the agent could not recover from. ERROR is logged. The run halts unless retry is configured.
PENDING → SKIPPED — the agent decided the task is no longer relevant (preconditions changed, parent task failed). DECISION is logged.
Any state → IN_PROGRESS — the operator can manually re-run a task from the dashboard.

Setting up Auto Mode

Define a goal from the dashboard or via the /goal chat command. Give it a name and a description.
Plan tasks — let the agent decompose the goal, or provide the task list manually.
Create a schedule in Scheduler with a cron expression. Attach it to the goal.
The agent executes tasks on each trigger. Progress is visible in the diary and in Sessions.
Inspect results via /status in chat, or the dashboard. Adjust the goal, tasks, or schedule as you learn.

Goals and tasks persist in workspace/auto-mode/ and survive container restarts. A paused goal resumes on the next scheduled tick.

What to do next

Scheduled code review

End-to-end recipe: build a daily code review goal with a custom skill and a cron schedule.

Cookbook

Delayed actions

For one-off future actions rather than recurring workflows, use delayed actions instead.

User Guide

Delayed Actions

How the runtime owns delivery timing and safety re-evaluation for deferred agent actions.

User Guide

Dashboard

Navigate the dashboard: Chat, Settings, Scheduler, Sessions, Logs, Skills, Diagnostics, and the Plugin Marketplace.

Cookbook

Scheduled Code Review

Set up a recurring daily code review using Auto Mode, a custom skill, and a cron schedule. Verify the first run and iterate.